Decision methodology of end-milling conditions using data-mining

The purpose of the present study is to apply data-mining methods to support the decision of reasonable cutting conditions. Although an enormous amount of information is listed in a catalog, it is not possible to know all of it. Seen from the viewpoint of the user, this enormous amount of information becomes a hindrance. For example, even if an expert worker does not look at a catalog, in end-mill processing, he can decide the appropriate processing condition efficiently from experience; however, this type of situation creates difficult problems for an unskilled worker or a skilled worker with little experience. The recommended cutting condition for every type of material is listed in a catalog together with the appropriate tool, but it takes much time and labor to search and examine the catalog to find the right tool, and this process is inefficient. The main subject of our research was to support the processing condition of the end-mill for each precision tool efficiently based on end-mill clusters. Our research applied the techniques of data mining, in particular, non-hierarchy clustering and hierarchy clustering, to catalog data. With these techniques, we applied multiple regression analysis and reached the following main conclusions. As a first step, we paid attention to the shape element of catalog data. In addition to using conventional mining processes, we grouped end-mills from the viewpoint of tool shape, which meant the ratio of dimensions, visually by applying the K-means method. We applied variable cluster analysis next to each cluster and extracted an predictor variable to represent each cluster, and we performed multiple regression analysis and derived a cutting condition decision formula. The cutting condition decision formula provided high accuracy. The accuracy was higher than the results achieved through mining of all data. A more highly precise processing condition decision formula was derived by doing mining again, excluding the peculiar data clusters such as small diameter end-mill. We understood what was effective for cutting condition decision to be factors related to blade length and the ratio of the full length, factors which have not been singled out through background knowledge or expert knowledge, but were noticed as an effect of catalog mining.